X-Git-Url: https://git.arvados.org/arvados.git/blobdiff_plain/de3ced158ccf579efbe9bfe0e23e03eb9417daeb..ae92d144610446849eb568247a44f02ae985c281:/sdk/cwl/arvados_cwl/executor.py diff --git a/sdk/cwl/arvados_cwl/executor.py b/sdk/cwl/arvados_cwl/executor.py index 190ea35e2a..680ca0b7b2 100644 --- a/sdk/cwl/arvados_cwl/executor.py +++ b/sdk/cwl/arvados_cwl/executor.py @@ -18,6 +18,7 @@ import json import re from functools import partial import time +import urllib from cwltool.errors import WorkflowException import cwltool.workflow @@ -31,19 +32,18 @@ from arvados.errors import ApiError import arvados_cwl.util from .arvcontainer import RunnerContainer -from .arvjob import RunnerJob, RunnerTemplate from .runner import Runner, upload_docker, upload_job_order, upload_workflow_deps from .arvtool import ArvadosCommandTool, validate_cluster_target, ArvadosExpressionTool from .arvworkflow import ArvadosWorkflow, upload_workflow from .fsaccess import CollectionFsAccess, CollectionFetcher, collectionResolver, CollectionCache, pdh_size from .perf import Perf from .pathmapper import NoFollowPathMapper -from .task_queue import TaskQueue +from cwltool.task_queue import TaskQueue from .context import ArvLoadingContext, ArvRuntimeContext from ._version import __version__ from cwltool.process import shortname, UnsupportedRequirement, use_custom_schema -from cwltool.pathmapper import adjustFileObjs, adjustDirObjs, get_listing, visit_class +from cwltool.utils import adjustFileObjs, adjustDirObjs, get_listing, visit_class, aslist from cwltool.command_line_tool import compute_checksums from cwltool.load_tool import load_tool @@ -91,8 +91,8 @@ class RuntimeStatusLoggingHandler(logging.Handler): class ArvCwlExecutor(object): - """Execute a CWL tool or workflow, submit work (using either jobs or - containers API), wait for them to complete, and report output. + """Execute a CWL tool or workflow, submit work (using containers API), + wait for them to complete, and report output. """ @@ -100,7 +100,8 @@ class ArvCwlExecutor(object): arvargs=None, keep_client=None, num_retries=4, - thread_count=4): + thread_count=4, + stdout=sys.stdout): if arvargs is None: arvargs = argparse.Namespace() @@ -133,6 +134,7 @@ class ArvCwlExecutor(object): self.should_estimate_cache_size = True self.fs_access = None self.secret_store = None + self.stdout = stdout if keep_client is not None: self.keep_client = keep_client @@ -154,7 +156,7 @@ class ArvCwlExecutor(object): num_retries=self.num_retries) self.work_api = None - expected_api = ["containers", "jobs"] + expected_api = ["containers"] for api in expected_api: try: methods = self.api._rootDesc.get('resources')[api]['methods'] @@ -172,19 +174,11 @@ class ArvCwlExecutor(object): raise Exception("Unsupported API '%s', expected one of %s" % (arvargs.work_api, expected_api)) if self.work_api == "jobs": - logger.warning(""" + logger.error(""" ******************************* -Using the deprecated 'jobs' API. - -To get rid of this warning: - -Users: read about migrating at -http://doc.arvados.org/user/cwl/cwl-style.html#migrate -and use the option --api=containers - -Admins: configure the cluster to disable the 'jobs' API as described at: -http://doc.arvados.org/install/install-api-server.html#disable_api_methods +The 'jobs' API is no longer supported. *******************************""") + exit(1) self.loadingContext = ArvLoadingContext(vars(arvargs)) self.loadingContext.fetcher_constructor = self.fetcher_constructor @@ -259,50 +253,37 @@ http://doc.arvados.org/install/install-api-server.html#disable_api_methods activity statuses, for example in the RuntimeStatusLoggingHandler. """ with self.workflow_eval_lock: - current = arvados_cwl.util.get_current_container(self.api, self.num_retries, logger) + current = None + try: + current = arvados_cwl.util.get_current_container(self.api, self.num_retries, logger) + except Exception as e: + logger.info("Couldn't get current container: %s", e) if current is None: return runtime_status = current.get('runtime_status', {}) - # In case of status being an error, only report the first one. - if kind == 'error': - if not runtime_status.get('error'): - runtime_status.update({ - 'error': message - }) - if detail is not None: - runtime_status.update({ - 'errorDetail': detail - }) - # Further errors are only mentioned as a count. - else: - # Get anything before an optional 'and N more' string. - try: - error_msg = re.match( - r'^(.*?)(?=\s*\(and \d+ more\)|$)', runtime_status.get('error')).groups()[0] - more_failures = re.match( - r'.*\(and (\d+) more\)', runtime_status.get('error')) - except TypeError: - # Ignore tests stubbing errors - return - if more_failures: - failure_qty = int(more_failures.groups()[0]) - runtime_status.update({ - 'error': "%s (and %d more)" % (error_msg, failure_qty+1) - }) - else: - runtime_status.update({ - 'error': "%s (and 1 more)" % error_msg - }) - elif kind in ['warning', 'activity']: - # Record the last warning/activity status without regard of - # previous occurences. + if kind in ('error', 'warning'): + updatemessage = runtime_status.get(kind, "") + if not updatemessage: + updatemessage = message + + # Subsequent messages tacked on in detail + updatedetail = runtime_status.get(kind+'Detail', "") + maxlines = 40 + if updatedetail.count("\n") < maxlines: + if updatedetail: + updatedetail += "\n" + updatedetail += message + "\n" + + if detail: + updatedetail += detail + "\n" + + if updatedetail.count("\n") >= maxlines: + updatedetail += "\nSome messages may have been omitted. Check the full log." + runtime_status.update({ - kind: message + kind: updatemessage, + kind+'Detail': updatedetail, }) - if detail is not None: - runtime_status.update({ - kind+"Detail": detail - }) else: # Ignore any other status kind return @@ -339,7 +320,7 @@ http://doc.arvados.org/install/install-api-server.html#disable_api_methods return "[%s %s]" % (self.work_api[0:-1], obj.name) def poll_states(self): - """Poll status of jobs or containers listed in the processes dict. + """Poll status of containers listed in the processes dict. Runs in a separate thread. """ @@ -360,8 +341,6 @@ http://doc.arvados.org/install/install-api-server.html#disable_api_methods begin_poll = time.time() if self.work_api == "containers": table = self.poll_api.container_requests() - elif self.work_api == "jobs": - table = self.poll_api.jobs() pageSize = self.poll_api._rootDesc.get('maxItemsPerResponse', 1000) @@ -410,18 +389,11 @@ http://doc.arvados.org/install/install-api-server.html#disable_api_methods def check_features(self, obj, parentfield=""): if isinstance(obj, dict): - if obj.get("writable") and self.work_api != "containers": - raise SourceLine(obj, "writable", UnsupportedRequirement).makeError("InitialWorkDir feature 'writable: true' not supported with --api=jobs") if obj.get("class") == "DockerRequirement": if obj.get("dockerOutputDirectory"): - if self.work_api != "containers": - raise SourceLine(obj, "dockerOutputDirectory", UnsupportedRequirement).makeError( - "Option 'dockerOutputDirectory' of DockerRequirement not supported with --api=jobs.") if not obj.get("dockerOutputDirectory").startswith('/'): raise SourceLine(obj, "dockerOutputDirectory", validate.ValidationException).makeError( "Option 'dockerOutputDirectory' must be an absolute path.") - if obj.get("class") == "http://commonwl.org/cwltool#Secrets" and self.work_api != "containers": - raise SourceLine(obj, "class", UnsupportedRequirement).makeError("Secrets not supported with --api=jobs") if obj.get("class") == "InplaceUpdateRequirement": if obj["inplaceUpdate"] and parentfield == "requirements": raise SourceLine(obj, "class", UnsupportedRequirement).makeError("InplaceUpdateRequirement not supported for keep collections.") @@ -462,7 +434,7 @@ http://doc.arvados.org/install/install-api-server.html#disable_api_methods srccollection = sp[0][5:] try: reader = self.collection_cache.get(srccollection) - srcpath = "/".join(sp[1:]) if len(sp) > 1 else "." + srcpath = urllib.parse.unquote("/".join(sp[1:]) if len(sp) > 1 else ".") final.copy(srcpath, v.target, source_collection=reader, overwrite=False) except arvados.errors.ArgumentError as e: logger.error("Creating CollectionReader for '%s' '%s': %s", k, v, e) @@ -521,14 +493,7 @@ http://doc.arvados.org/install/install-api-server.html#disable_api_methods }).execute(num_retries=self.num_retries) except Exception: logger.exception("Setting container output") - return - elif self.work_api == "jobs" and "TASK_UUID" in os.environ: - self.api.job_tasks().update(uuid=os.environ["TASK_UUID"], - body={ - 'output': self.final_output_collection.portable_data_hash(), - 'success': self.final_status == "success", - 'progress':1.0 - }).execute(num_retries=self.num_retries) + raise def apply_reqs(self, job_order_object, tool): if "https://w3id.org/cwl/cwl#requirements" in job_order_object: @@ -542,10 +507,15 @@ http://doc.arvados.org/install/install-api-server.html#disable_api_methods for req in job_reqs: tool.requirements.append(req) - def arv_executor(self, tool, job_order, runtimeContext, logger=None): + def arv_executor(self, updated_tool, job_order, runtimeContext, logger=None): self.debug = runtimeContext.debug - tool.visit(self.check_features) + workbench1 = self.api.config()["Services"]["Workbench1"]["ExternalURL"] + workbench2 = self.api.config()["Services"]["Workbench2"]["ExternalURL"] + controller = self.api.config()["Services"]["Controller"]["ExternalURL"] + logger.info("Using cluster %s (%s)", self.api.config()["ClusterID"], workbench2 or workbench1 or controller) + + updated_tool.visit(self.check_features) self.project_uuid = runtimeContext.project_uuid self.pipeline = None @@ -565,28 +535,41 @@ http://doc.arvados.org/install/install-api-server.html#disable_api_methods if runtimeContext.submit_request_uuid and self.work_api != "containers": raise Exception("--submit-request-uuid requires containers API, but using '{}' api".format(self.work_api)) + default_storage_classes = ",".join([k for k,v in self.api.config().get("StorageClasses", {"default": {"Default": True}}).items() if v.get("Default") is True]) + if runtimeContext.storage_classes == "default": + runtimeContext.storage_classes = default_storage_classes + if runtimeContext.intermediate_storage_classes == "default": + runtimeContext.intermediate_storage_classes = default_storage_classes + if not runtimeContext.name: - runtimeContext.name = self.name = tool.tool.get("label") or tool.metadata.get("label") or os.path.basename(tool.tool["id"]) + runtimeContext.name = self.name = updated_tool.tool.get("label") or updated_tool.metadata.get("label") or os.path.basename(updated_tool.tool["id"]) # Upload local file references in the job order. job_order = upload_job_order(self, "%s input" % runtimeContext.name, - tool, job_order) + updated_tool, job_order) + + # the last clause means: if it is a command line tool, and we + # are going to wait for the result, and always_submit_runner + # is false, then we don't submit a runner process. submitting = (runtimeContext.update_workflow or runtimeContext.create_workflow or (runtimeContext.submit and not - (tool.tool["class"] == "CommandLineTool" and + (updated_tool.tool["class"] == "CommandLineTool" and runtimeContext.wait and not runtimeContext.always_submit_runner))) loadingContext = self.loadingContext.copy() loadingContext.do_validate = False - loadingContext.do_update = False if submitting: + loadingContext.do_update = False # Document may have been auto-updated. Reload the original # document with updating disabled because we want to - # submit the original document, not the auto-updated one. - tool = load_tool(tool.tool["id"], loadingContext) + # submit the document with its original CWL version, not + # the auto-updated one. + tool = load_tool(updated_tool.tool["id"], loadingContext) + else: + tool = updated_tool # Upload direct dependencies of workflow steps, get back mapping of files to keep references. # Also uploads docker images. @@ -604,25 +587,16 @@ http://doc.arvados.org/install/install-api-server.html#disable_api_methods existing_uuid = runtimeContext.update_workflow if existing_uuid or runtimeContext.create_workflow: # Create a pipeline template or workflow record and exit. - if self.work_api == "jobs": - tmpl = RunnerTemplate(self, tool, job_order, - runtimeContext.enable_reuse, - uuid=existing_uuid, - submit_runner_ram=runtimeContext.submit_runner_ram, - name=runtimeContext.name, - merged_map=merged_map, - loadingContext=loadingContext) - tmpl.save() - # cwltool.main will write our return value to stdout. - return (tmpl.uuid, "success") - elif self.work_api == "containers": - return (upload_workflow(self, tool, job_order, + if self.work_api == "containers": + uuid = upload_workflow(self, tool, job_order, self.project_uuid, uuid=existing_uuid, submit_runner_ram=runtimeContext.submit_runner_ram, name=runtimeContext.name, - merged_map=merged_map), - "success") + merged_map=merged_map, + submit_runner_image=runtimeContext.submit_runner_image) + self.stdout.write(uuid + "\n") + return (None, "success") self.apply_reqs(job_order, tool) @@ -641,12 +615,6 @@ http://doc.arvados.org/install/install-api-server.html#disable_api_methods runtimeContext.docker_outdir = "/var/spool/cwl" runtimeContext.tmpdir = "/tmp" runtimeContext.docker_tmpdir = "/tmp" - elif self.work_api == "jobs": - if runtimeContext.priority != DEFAULT_PRIORITY: - raise Exception("--priority not implemented for jobs API.") - runtimeContext.outdir = "$(task.outdir)" - runtimeContext.docker_outdir = "$(task.outdir)" - runtimeContext.tmpdir = "$(task.tmpdir)" if runtimeContext.priority < 1 or runtimeContext.priority > 1000: raise Exception("--priority must be in the range 1..1000.") @@ -670,40 +638,23 @@ http://doc.arvados.org/install/install-api-server.html#disable_api_methods if runtimeContext.submit: # Submit a runner job to run the workflow for us. if self.work_api == "containers": - if tool.tool["class"] == "CommandLineTool" and runtimeContext.wait and (not runtimeContext.always_submit_runner): - runtimeContext.runnerjob = tool.tool["id"] + if submitting: + tool = RunnerContainer(self, updated_tool, + tool, loadingContext, runtimeContext.enable_reuse, + self.output_name, + self.output_tags, + submit_runner_ram=runtimeContext.submit_runner_ram, + name=runtimeContext.name, + on_error=runtimeContext.on_error, + submit_runner_image=runtimeContext.submit_runner_image, + intermediate_output_ttl=runtimeContext.intermediate_output_ttl, + merged_map=merged_map, + priority=runtimeContext.priority, + secret_store=self.secret_store, + collection_cache_size=runtimeContext.collection_cache_size, + collection_cache_is_default=self.should_estimate_cache_size) else: - tool = RunnerContainer(self, tool, loadingContext, runtimeContext.enable_reuse, - self.output_name, - self.output_tags, - submit_runner_ram=runtimeContext.submit_runner_ram, - name=runtimeContext.name, - on_error=runtimeContext.on_error, - submit_runner_image=runtimeContext.submit_runner_image, - intermediate_output_ttl=runtimeContext.intermediate_output_ttl, - merged_map=merged_map, - priority=runtimeContext.priority, - secret_store=self.secret_store, - collection_cache_size=runtimeContext.collection_cache_size, - collection_cache_is_default=self.should_estimate_cache_size) - elif self.work_api == "jobs": - tool = RunnerJob(self, tool, loadingContext, runtimeContext.enable_reuse, - self.output_name, - self.output_tags, - submit_runner_ram=runtimeContext.submit_runner_ram, - name=runtimeContext.name, - on_error=runtimeContext.on_error, - submit_runner_image=runtimeContext.submit_runner_image, - merged_map=merged_map) - elif runtimeContext.cwl_runner_job is None and self.work_api == "jobs": - # Create pipeline for local run - self.pipeline = self.api.pipeline_instances().create( - body={ - "owner_uuid": self.project_uuid, - "name": runtimeContext.name if runtimeContext.name else shortname(tool.tool["id"]), - "components": {}, - "state": "RunningOnClient"}).execute(num_retries=self.num_retries) - logger.info("Pipeline instance %s", self.pipeline["uuid"]) + runtimeContext.runnerjob = tool.tool["id"] if runtimeContext.cwl_runner_job is not None: self.uuid = runtimeContext.cwl_runner_job.get('uuid') @@ -715,7 +666,8 @@ http://doc.arvados.org/install/install-api-server.html#disable_api_methods if runtimeContext.submit and not runtimeContext.wait: runnerjob = next(jobiter) runnerjob.run(runtimeContext) - return (runnerjob.uuid, "success") + self.stdout.write(runnerjob.uuid+"\n") + return (None, "success") current_container = arvados_cwl.util.get_current_container(self.api, self.num_retries, logger) if current_container: @@ -780,11 +732,16 @@ http://doc.arvados.org/install/install-api-server.html#disable_api_methods if self.pipeline: self.api.pipeline_instances().update(uuid=self.pipeline["uuid"], body={"state": "Failed"}).execute(num_retries=self.num_retries) - if runtimeContext.submit and isinstance(tool, Runner): - runnerjob = tool - if runnerjob.uuid and self.work_api == "containers": - self.api.container_requests().update(uuid=runnerjob.uuid, - body={"priority": "0"}).execute(num_retries=self.num_retries) + + if self.work_api == "containers" and not current_container: + # Not running in a crunch container, so cancel any outstanding processes. + for p in self.processes: + try: + self.api.container_requests().update(uuid=p, + body={"priority": "0"} + ).execute(num_retries=self.num_retries) + except Exception: + pass finally: self.workflow_eval_lock.release() self.task_queue.drain() @@ -800,13 +757,21 @@ http://doc.arvados.org/install/install-api-server.html#disable_api_methods if runtimeContext.submit and isinstance(tool, Runner): logger.info("Final output collection %s", tool.final_output) + if workbench2 or workbench1: + logger.info("Output at %scollections/%s", workbench2 or workbench1, tool.final_output) else: if self.output_name is None: self.output_name = "Output of %s" % (shortname(tool.tool["id"])) if self.output_tags is None: self.output_tags = "" - storage_classes = runtimeContext.storage_classes.strip().split(",") + storage_classes = "" + storage_class_req, _ = tool.get_requirement("http://arvados.org/cwl#OutputStorageClass") + if storage_class_req and storage_class_req.get("finalStorageClass"): + storage_classes = aslist(storage_class_req["finalStorageClass"]) + else: + storage_classes = runtimeContext.storage_classes.strip().split(",") + self.final_output, self.final_output_collection = self.make_output_collection(self.output_name, storage_classes, self.output_tags, self.final_output) self.set_crunch_output()